classic_pythia-2.8b-step2000/groups/raft.json

[
  {
    "title": "RAFT (Real-world Annotated Few-Shot)",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "markdown": false
        },
        {
          "value": 0.41590909090909095,
          "description": "min=0.025, mean=0.416, max=0.975, sum=4.575 (11)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.2537619778871928,
          "description": "min=0.098, mean=0.254, max=0.427, sum=2.791 (11)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.3568181818181818,
          "description": "min=0, mean=0.357, max=0.975, sum=3.925 (11)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.38181818181818183,
          "description": "min=0, mean=0.382, max=0.975, sum=4.2 (11)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "description": "11 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "11 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=440 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 4.6045454545454545,
          "description": "min=0.7, mean=4.605, max=5, sum=50.65 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 869.6909090909089,
          "description": "min=280.35, mean=869.691, max=1756.575, sum=9566.6 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 4.661363636363637,
          "description": "min=1.1, mean=4.661, max=18.65, sum=51.275 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=11 (11)",
          "style": {},
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft.json"
      }
    ],
    "name": "raft"
  },
  {
    "title": "subset: ade_corpus_v2",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20ade_corpus_v2&runSpecs=%5B%22raft%3Asubset%3Dade_corpus_v2%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=ade_corpus_v2,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.35,
          "description": "min=0.35, mean=0.35, max=0.35, sum=0.35 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.2200087424584905,
          "description": "min=0.22, mean=0.22, max=0.22, sum=0.22 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.3,
          "description": "min=0.3, mean=0.3, max=0.3, sum=0.3 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.25,
          "description": "min=0.25, mean=0.25, max=0.25, sum=0.25 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 382.625,
          "description": "min=382.625, mean=382.625, max=382.625, sum=382.625 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.175,
          "description": "min=5.175, mean=5.175, max=5.175, sum=5.175 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:ade_corpus_v2.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:ade_corpus_v2.json"
      }
    ],
    "name": "raft_subset:ade_corpus_v2"
  },
  {
    "title": "subset: banking_77",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20banking_77&runSpecs=%5B%22raft%3Asubset%3Dbanking_77%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=banking_77,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.05,
          "description": "min=0.05, mean=0.05, max=0.05, sum=0.05 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.09826959298818556,
          "description": "min=0.098, mean=0.098, max=0.098, sum=0.098 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.025,
          "description": "min=0.025, mean=0.025, max=0.025, sum=0.025 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.025,
          "description": "min=0.025, mean=0.025, max=0.025, sum=0.025 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 855.2,
          "description": "min=855.2, mean=855.2, max=855.2, sum=855.2 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 3.625,
          "description": "min=3.625, mean=3.625, max=3.625, sum=3.625 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:banking_77.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:banking_77.json"
      }
    ],
    "name": "raft_subset:banking_77"
  },
  {
    "title": "subset: neurips_impact_statement_risks",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20neurips_impact_statement_risks&runSpecs=%5B%22raft%3Asubset%3Dneurips_impact_statement_risks%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=neurips_impact_statement_risks,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.4,
          "description": "min=0.4, mean=0.4, max=0.4, sum=0.4 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.3102733709723004,
          "description": "min=0.31, mean=0.31, max=0.31, sum=0.31 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.175,
          "description": "min=0.175, mean=0.175, max=0.175, sum=0.175 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.275,
          "description": "min=0.275, mean=0.275, max=0.275, sum=0.275 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 4.95,
          "description": "min=4.95, mean=4.95, max=4.95, sum=4.95 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1459.25,
          "description": "min=1459.25, mean=1459.25, max=1459.25, sum=1459.25 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.3,
          "description": "min=5.3, mean=5.3, max=5.3, sum=5.3 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:neurips_impact_statement_risks.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:neurips_impact_statement_risks.json"
      }
    ],
    "name": "raft_subset:neurips_impact_statement_risks"
  },
  {
    "title": "subset: one_stop_english",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20one_stop_english&runSpecs=%5B%22raft%3Asubset%3Done_stop_english%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=one_stop_english,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.025,
          "description": "min=0.025, mean=0.025, max=0.025, sum=0.025 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.20856635850640312,
          "description": "min=0.209, mean=0.209, max=0.209, sum=0.209 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.7,
          "description": "min=0.7, mean=0.7, max=0.7, sum=0.7 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1594.6,
          "description": "min=1594.6, mean=1594.6, max=1594.6, sum=1594.6 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 18.65,
          "description": "min=18.65, mean=18.65, max=18.65, sum=18.65 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:one_stop_english.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:one_stop_english.json"
      }
    ],
    "name": "raft_subset:one_stop_english"
  },
  {
    "title": "subset: overruling",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20overruling&runSpecs=%5B%22raft%3Asubset%3Doverruling%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=overruling,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.425,
          "description": "min=0.425, mean=0.425, max=0.425, sum=0.425 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.23160871290153412,
          "description": "min=0.232, mean=0.232, max=0.232, sum=0.232 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.425,
          "description": "min=0.425, mean=0.425, max=0.425, sum=0.425 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.425,
          "description": "min=0.425, mean=0.425, max=0.425, sum=0.425 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 301.025,
          "description": "min=301.025, mean=301.025, max=301.025, sum=301.025 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 4.1,
          "description": "min=4.1, mean=4.1, max=4.1, sum=4.1 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:overruling.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:overruling.json"
      }
    ],
    "name": "raft_subset:overruling"
  },
  {
    "title": "subset: semiconductor_org_types",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20semiconductor_org_types&runSpecs=%5B%22raft%3Asubset%3Dsemiconductor_org_types%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=semiconductor_org_types,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.1,
          "description": "min=0.1, mean=0.1, max=0.1, sum=0.1 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.42719035636028496,
          "description": "min=0.427, mean=0.427, max=0.427, sum=0.427 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.075,
          "description": "min=0.075, mean=0.075, max=0.075, sum=0.075 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.1,
          "description": "min=0.1, mean=0.1, max=0.1, sum=0.1 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 400.95,
          "description": "min=400.95, mean=400.95, max=400.95, sum=400.95 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.1,
          "description": "min=1.1, mean=1.1, max=1.1, sum=1.1 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:semiconductor_org_types.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:semiconductor_org_types.json"
      }
    ],
    "name": "raft_subset:semiconductor_org_types"
  },
  {
    "title": "subset: systematic_review_inclusion",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20systematic_review_inclusion&runSpecs=%5B%22raft%3Asubset%3Dsystematic_review_inclusion%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=systematic_review_inclusion,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.975,
          "description": "min=0.975, mean=0.975, max=0.975, sum=0.975 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.15178136408014306,
          "description": "min=0.152, mean=0.152, max=0.152, sum=0.152 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.975,
          "description": "min=0.975, mean=0.975, max=0.975, sum=0.975 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.975,
          "description": "min=0.975, mean=0.975, max=0.975, sum=0.975 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1349.025,
          "description": "min=1349.025, mean=1349.025, max=1349.025, sum=1349.025 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 2.0,
          "description": "min=2, mean=2, max=2, sum=2 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:systematic_review_inclusion.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:systematic_review_inclusion.json"
      }
    ],
    "name": "raft_subset:systematic_review_inclusion"
  },
  {
    "title": "subset: tai_safety_research",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20tai_safety_research&runSpecs=%5B%22raft%3Asubset%3Dtai_safety_research%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=tai_safety_research,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.5,
          "description": "min=0.5, mean=0.5, max=0.5, sum=0.5 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.2984436288043698,
          "description": "min=0.298, mean=0.298, max=0.298, sum=0.298 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.5,
          "description": "min=0.5, mean=0.5, max=0.5, sum=0.5 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.425,
          "description": "min=0.425, mean=0.425, max=0.425, sum=0.425 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1756.575,
          "description": "min=1756.575, mean=1756.575, max=1756.575, sum=1756.575 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 4.0,
          "description": "min=4, mean=4, max=4, sum=4 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:tai_safety_research.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:tai_safety_research.json"
      }
    ],
    "name": "raft_subset:tai_safety_research"
  },
  {
    "title": "subset: terms_of_service",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20terms_of_service&runSpecs=%5B%22raft%3Asubset%3Dterms_of_service%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=terms_of_service,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.525,
          "description": "min=0.525, mean=0.525, max=0.525, sum=0.525 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.2749345590427505,
          "description": "min=0.275, mean=0.275, max=0.275, sum=0.275 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.4,
          "description": "min=0.4, mean=0.4, max=0.4, sum=0.4 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.5,
          "description": "min=0.5, mean=0.5, max=0.5, sum=0.5 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 820.575,
          "description": "min=820.575, mean=820.575, max=820.575, sum=820.575 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 2.35,
          "description": "min=2.35, mean=2.35, max=2.35, sum=2.35 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:terms_of_service.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:terms_of_service.json"
      }
    ],
    "name": "raft_subset:terms_of_service"
  },
  {
    "title": "subset: tweet_eval_hate",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20tweet_eval_hate&runSpecs=%5B%22raft%3Asubset%3Dtweet_eval_hate%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=tweet_eval_hate,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.525,
          "description": "min=0.525, mean=0.525, max=0.525, sum=0.525 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.1863503806886519,
          "description": "min=0.186, mean=0.186, max=0.186, sum=0.186 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.525,
          "description": "min=0.525, mean=0.525, max=0.525, sum=0.525 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.525,
          "description": "min=0.525, mean=0.525, max=0.525, sum=0.525 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 366.425,
          "description": "min=366.425, mean=366.425, max=366.425, sum=366.425 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 2.975,
          "description": "min=2.975, mean=2.975, max=2.975, sum=2.975 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:tweet_eval_hate.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:tweet_eval_hate.json"
      }
    ],
    "name": "raft_subset:tweet_eval_hate"
  },
  {
    "title": "subset: twitter_complaints",
    "header": [
      {
        "value": "Model",
        "markdown": false,
        "metadata": {}
      },
      {
        "value": "EM",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT"
        }
      },
      {
        "value": "ECE (10-bin)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n10-bin expected calibration error: The average difference between the model's confidence and accuracy, averaged across 10 bins where each bin contains an equal number of points (only computed for classification tasks). Warning - not reliable for small datasets (e.g., with < 300 examples) because each bin will have very few examples.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "ECE (10-bin)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "EM (Robustness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Robustness"
        }
      },
      {
        "value": "EM (Fairness)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nQuasi-exact match: Fraction of instances that the predicted output matches a correct reference up to light processing.\n- Perturbation Fairness: Computes worst case over different fairness perturbations (changing dialect, race of names, gender).",
        "markdown": false,
        "lower_is_better": false,
        "metadata": {
          "metric": "EM",
          "run_group": "RAFT",
          "perturbation": "Fairness"
        }
      },
      {
        "value": "Stereotypes (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (race, profession): Measures uneven association of racial groups (Asian, Hispanic, White) with target professions. This measurement is based on cooccurence statistics between the racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Stereotypes (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nStereotypical associations (gender, profession): Measures uneven association of gender groups (male, female) with target professions. This measurement is based on cooccurence statistics between the gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)) and the target professions (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Stereotypes (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (race)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (race): Measures uneven representation of racial groups (Asian, Hispanic, White). This measurement is based on disparities in the frequency statistics across racially-associated names (word list from [Garg et al., 2018](https://www.pnas.org/doi/10.1073/pnas.1720347115); race associations based on US Census statistics).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (race)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Representation (gender)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDemographic representation (gender): Measures uneven representation of gender groups (male, female). This measurement is based on disparities in the frequency statistics across gender terms (word list from [Bolukbasi et al., 2016](https://papers.nips.cc/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html)).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Representation (gender)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Toxic fraction",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nToxic fraction: Fraction of model outputs that are toxic (based on the PerspectiveAPI toxicity classifier).",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Toxic fraction",
          "run_group": "RAFT"
        }
      },
      {
        "value": "Denoised inference time (s)",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\nDenoised inference runtime (s): Average time to process a request to the model minus performance contention by using profiled runtimes from multiple trials of SyntheticEfficiencyScenario.",
        "markdown": false,
        "lower_is_better": true,
        "metadata": {
          "metric": "Denoised inference time (s)",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# eval",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# eval: Number of evaluation instances.",
        "markdown": false,
        "metadata": {
          "metric": "# eval",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# train",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# train: Number of training instances (e.g., in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "# train",
          "run_group": "RAFT"
        }
      },
      {
        "value": "truncated",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\ntruncated: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).",
        "markdown": false,
        "metadata": {
          "metric": "truncated",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# prompt tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# prompt tokens: Number of tokens in the prompt.",
        "markdown": false,
        "metadata": {
          "metric": "# prompt tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# output tokens",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# output tokens: Actual number of output tokens.",
        "markdown": false,
        "metadata": {
          "metric": "# output tokens",
          "run_group": "RAFT"
        }
      },
      {
        "value": "# trials",
        "description": "The Real-world annotated few-shot (RAFT) meta-benchmark of 11 real-world text classification tasks [(Alex et al., 2021)](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/ca46c1b9512a7a8315fa3c5a946e8265-Abstract-round2.html).\n\n# trials: Number of trials, where in each trial we choose an independent, random set of training instances.",
        "markdown": false,
        "metadata": {
          "metric": "# trials",
          "run_group": "RAFT"
        }
      }
    ],
    "rows": [
      [
        {
          "value": "EleutherAI/pythia-2.8b",
          "description": "",
          "href": "?group=raft&subgroup=subset%3A%20twitter_complaints&runSpecs=%5B%22raft%3Asubset%3Dtwitter_complaints%2Cmodel%3DEleutherAI_pythia-2.8b%2Cdata_augmentation%3Dcanonical%22%5D",
          "markdown": false,
          "run_spec_names": [
            "raft:subset=twitter_complaints,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
          ]
        },
        {
          "value": 0.7,
          "description": "min=0.7, mean=0.7, max=0.7, sum=0.7 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.38395468995600673,
          "description": "min=0.384, mean=0.384, max=0.384, sum=0.384 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.525,
          "description": "min=0.525, mean=0.525, max=0.525, sum=0.525 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "value": 0.7,
          "description": "min=0.7, mean=0.7, max=0.7, sum=0.7 (1)",
          "style": {
            "font-weight": "bold"
          },
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "(0)",
          "style": {},
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "description": "1 matching runs, but no matching metrics",
          "markdown": false
        },
        {
          "value": 40.0,
          "description": "min=40, mean=40, max=40, sum=40 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 5.0,
          "description": "min=5, mean=5, max=5, sum=5 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 0.0,
          "description": "min=0, mean=0, max=0, sum=0 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 280.35,
          "description": "min=280.35, mean=280.35, max=280.35, sum=280.35 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 2.0,
          "description": "min=2, mean=2, max=2, sum=2 (1)",
          "style": {},
          "markdown": false
        },
        {
          "value": 1.0,
          "description": "min=1, mean=1, max=1, sum=1 (1)",
          "style": {},
          "markdown": false
        }
      ]
    ],
    "links": [
      {
        "text": "LaTeX",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/raft_raft_subset:twitter_complaints.tex"
      },
      {
        "text": "JSON",
        "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/raft_raft_subset:twitter_complaints.json"
      }
    ],
    "name": "raft_subset:twitter_complaints"
  }
]